Imbalanced enterprise credit evaluation with DTE-SBD: Decision tree ensemble based on SMOTE and bagging with differentiated sampling rates
Imbalanced enterprise credit evaluation with DTE-SBD: Decision tree ensemble based on SMOTE and bagging with differentiated sampling rates
复制标题
基于DTE-SBD的不平衡企业信用评估：基于SMOTE和差异采样率bagging的决策树集成
DOI：
10.1016/j.ins.2017.10.017
复制
发表时间：
2018-01-01
影响因子：
8.1
通讯作者：
Li, Hui
中科院分区：
文献类型：
--
作者：
Sun, Jie;Lang, Jie;Li, Hui
企业信用评估模型是银行和企业风险管理的重要工具，但如何构建有效的决策树集成模型来进行非均衡企业信用评估的研究较少。本文提出了一种新的基于合成少数过采样技术（SMOTE）和差分采样率Bagging集成学习算法（DSR）的非均衡企业信用评估DT集成模型DTE-SBD（Decision Tree Ensemination based on SMOTE，Bagging and DSR）。在不同的迭代次数下进行基Di分类器训练，新的正SMOTE与DSR产生不同程度的（高风险）样本，并产生不同数量的阴性（低风险）样本的抽取采用Bagging with DSR替换，但在一定采样率的迭代同时，包括原始和新的训练正样本与抽取的训练负样本的数量相同，并且它们被组合以训练DT基分类器。因此，DTE-SBD不仅可以解决企业信用评估中的类不平衡问题，而且可以增加DT集成中基分类器的多样性。利用中国552家上市公司的财务数据进行了100次实证实验，比较了纯DT、过抽样DT、欠抽样DT、SMOTE DT、Bagging DT和DTE-SBD六种模型对非均衡企业信用评价的绩效。实验结果表明，DTE-SBD模型显著优于其他五种模型，对不平衡的企业信用评估是有效的。(C)2017爱思唯尔公司All rights reserved.
Enterprise credit evaluation model is an important tool for bank and enterprise risk management, but how to construct an effective decision tree (DT) ensemble model for imbalanced enterprise credit evaluation is seldom studied. This paper proposes a new DT ensemble model for imbalanced enterprise credit evaluation based on the synthetic minority over-sampling technique (SMOTE) and the Bagging ensemble learning algorithm with differentiated sampling rates (DSR), which is named as DTE-SBD (Decision Tree Ensemble based on SMOTE, Bagging and DSR). In different times of iteration for base Di classifier training, new positive (high risky) samples are produced to different degrees by SMOTE with DSR, and different numbers of negative (low risky) samples are drawn with replacement by Bagging with DSR However, in the same time of iteration with certain sampling rate, the training positive samples including the original and the new are of the same number as the drawn training negative samples, and they are combined to train a DT base classifier. Therefore, DTE-SBD can not only dispose the class imbalance problem of enterprise credit evaluation, but also increase the diversity of base classifiers for DT ensemble. Empirical experiment is carried out for 100 times with the financial data of 552 Chinese listed companies, and the performance of imbalanced enterprise credit evaluation is compared among the six models of pure DT, over-sampling DT, over-under-sampling DT, SMOTE DT, Bagging DT, and DTE-SBD. The experimental results indicate that DTE-SBD significantly outperforms the other five models and is effective for imbalanced enterprise credit evaluation. (C) 2017 Elsevier Inc. All rights reserved.